Internet connection

not significant, I wondered if latency would jump up/down as speed jumped down/up, e.g. to see if changes happen at whatever time farmer Giles turns on his milking parlour, or people in your robotics barn turn on the plasma cutter?

Pretty quiet here electrically. The Robot Centre/carpenters have a fair bit of 3 phase electrical machinery but much of my problem seems to happen overnight. I'll leave the PC running and see what the stats show tomorrow.

Sub zero temperatures overnight. Joints may pull apart.

I trow you still have a line fault :-)

I think you are correct. Clearly Open Reach are not going to install significant lengths of overhead cable in the present environment:-( No reason to not swap a pair they already have!

I have been monitoring the service with Pingplotter as suggested. During the last two nights the DSL has been recorded as *down* by my router stats: coinciding with a 1 second *no data* bar on PP. Each time the router has reconnected but at a reduced speed.

Currently 2883kB down SNR 31.4dB and 248kB up SNR 6.0dB

Plusnet are leaving the job open for the week.

Yes. If those stats are true, you are getting massive *bursts* of noise.

Did they switch you to a different DSLAM at the exchange?

If they did the only possible issue is a crappy line.

Last exchange job was a *lift and shift*.

Air rifle pellet? My misspent youth coming back to bite me!

There is an interaction between solar radiation and propagation of radio signals in the HF band. Typically the lower frequencies propagate better at night. Since ADSL2 uses frequency bands that approach the bottom end of the HF band, you see a greater influence from HF radio broadcasts overnight than during the day. Long overhead runs of telephone wire also make reasonably good HF "long wire" antenna. So a drop in SNR overnight is not uncommon.

Pingplotter shows huge *packet loss* issues. Beginning at 1.30am and ending at 7.40 am. gaps of up to several seconds. Ok now but speed around 3.5Mb only. Good enough for Usenet:-)

Minus 8 deg. C overnight here.

Don't tell Open Reach but I climbed a ladder where their cable runs through some trees to look for damage. I may have given their cables a tug or two:-)

Packet loss will happen as the noise floor rises. The FEC and interleaving can cope with some degradation of signal, but for any serious shift in line performance, data rate adaptation is the only real way to deal with it. Hence the slowing down.

Backward progress! Plusnet supplied a replacement router which appears better at re-connecting but in exchange for a reduction in speed. Currently around 1Mb down. I have tried a couple of resets without improvement.

My speed guarantee is 2.1Mb so I will start a fresh investigation Monday.

Latency (whatever that is) runs at around 30-mS apart from occasional steps up to 90mS or so.

The service interruptions last 1-2seconds before re-connection and generally overnight.

It is in your case almost the inverse of transfer speed.

The time it takes for a packet of data sent by your computer to reach its destination computer. This will be limited by how many routers it goes through, how fast the links are between them, and how congested those links are at the instant of transmission. Ultimately the speed of light comes into it, if your packet is going to NZ (12k miles) you wont get below 65 mSec or so.

The TCP/IP packets start out about 1500 bytes at the ethernet/wifi in the home, unfortunately in the case of Tim's ADSL connection, they then get sliced into 48 byte ATM cells, and on a dodgy line every one of those smaller cells has to make it over the copper (with retries if necessary) before the TCP packet has even made it to the local exchange, that's where the red bars on pingplotter will show up, if they don't in fact make it in time ...

In message snipped-for-privacy@mid.individual.net>, Tim Streater snipped-for-privacy@streater.me.uk> writes

Ah! That's clear enough. Ping plotter is pinging Google.

Hmm. Why would that appear to be worse overnight? I am hoping the upcoming gale will worsen the situation such that Openreach can find the fault

For similar reasons that interference on AM radio got worse at night ...

They could be swamped with calls and be diverted elsewhere for months :-(

That is not necessarily a bad thing. You can get more "conservative" firmware for a number of modems, that will tend to work with a larger target SNR. This does have the effect of lowering the maximum speed, but the trade off is more consistent and reliable connectivity.

However in your case 1Mbps seems like too much reduction - so you still have a proper fault somewhere.

Latency is the delay in packet transit - and particularly round trip time (aka "ping"). The spikes in time will be down to the impact of a burst error causing the loss of group of packets. More latency can make online interactions that require lots of short messages exchanged in sequence noticeably slower, and can also add lag to "real time" services lie VoIP. So low latency is in many cases a "good thing"

One of the techniques engineers use to improve reliability of comms systems is by adding Forward Error Correction (FEC) coding to the data being sent. This adds additional parity data to be added into the information sent, that allows the receiver to not only identify where corrupted bits were received, but also to correct them automatically without needing to request retransmission of the whole packet. This copes well with a random sprinkling of errors that flips a few bits spread over many bytes - and so will deal well with lines with background noise. It is unable however to deal with burst errors - i.e. a run of consecutive errors that wipe out a whole string of bits.

To cope with this, one can use interleaving. Typically splitting and spreading the bits of your FEC protected data out over time so that adjacent bits are not actually transmitted at the same time. Now if you get a burst error that wipes out a string of bits, it will ultimately cause a small amount of damage to lots of bytes, rather than the total destruction of a few bytes. The FEC can then fix these "less damaged" bytes.

The interruption will be when there are burst errors long enough that they exceed the interleaver depth, and the system can't deal with the problem using FEC. That then means that higher level protocols like TCP have to deal with it using Automatic Repeat reQuest (ARQ)- basically requesting retransmission of the complete lost packet again. In severe cases the higher level protocol can't fix it and enters a timeout or error state, and the connection is lost.

As I mentioned elsewhere in the thread, RF propagation changes at night (particularly due to changes in Ionospheric Propagation). That means that propagation of HF (and lower frequency) radio gets better at night:

Yup, might do...